Myocardial T1 mapping using an INSIL modeling and a Bayesian estimation method: Does improved cardiac T1 estimation accuracy and precision translate into enhanced clinical relevance in patients with chronic myocardial infarction? (RF_FR_344)

Friday, January 31, 2025

11:10 AM – 11:20 AM East Coast USA Time

Location: Blue Room PreFunction

Presenting Author(s)

Pierre Croisille, MD, PhD

Professor
University Hospital Saint-Etienne, France

Primary Author(s)

Magalie Viallon, PhD

Researcher
Univ Lyon, UJM-Saint-Etienne, INSA, CNRS UMR 5520, INSERM U1206, CREATIS, France

Co-Author(s)

TB

Timothé Boutelier, PhD

Olea Medical, France
HR

Habib Rebbah, PhD

R&I Engineer
Olea Medical, France
RP

Romain Pastre, MD

MD
University Hospital of Saint-Etienne, France

Background: Quantitative spin-lattice magnetization relaxation time (T1) mapping identifies differences between healthy and pathological tissues but the sensitivity for detecting abnormal T1 values and the subsequent clinical value of T1 related parameters, is dependent on the precision of T1 measures. In turn, it further conditions the capacity to define a pertinent threshold to stratify, monitor or grade disease. Hence, the quest for precise and accurate measurement has motivated efforts and led to continuous improvements. The objective of the current study is to evaluate the relevance and performance of most advanced methods against default T1 maps, and to assess the importance and significance of using more accurate T1 maps calculation.

Methods: T1 mapping MOLLI sequence was acquired on 73 acute myocardial infarct patients at 1.5 T. Classic InSiL technique and 2 implementations of recently proposed Bayesian estimation methods of the InSiL parameters, based on the same raw dataset as MOLLI, were calculated. Semi-automated segmentation of myocardial infarction (MI) and remote zone were performed and median T1 values and ECV of each method were compared.

Results: In native acquisitions, we found significant differences among methods both in MI (F (1,508, 223,1) = 72,42; p< 0.0001) and remote regions (F (1,972, 291,8) = 212,1; p< 0.0001). In MI regions, MOLLI vendor method lead to lower native T1 estimates except when compared to InSiL HBI(3p), with a mean difference of 34.9[4.8;64.9]ms (p=0.014). The largest mean T1 difference reached -91.3 [-104.1;-78.5] ms using InSiL

HBI(2p) (p=0< 0.001), while without significant differences when compared to InSiL LSQ(3p) (-2.9[--28.4;22.6]ms; p=0.90).

In remote regions, MOLLI vendor native T1 estimates were always below all the other approaches estimates.

The largest mean difference reached -100.2[-105.1;-95.2]ms when compared to InSiL LSQ(3p) and the smallest mean difference was when compared to InSiL HBI(3p) with -63.8[-78.7;-49]ms.

Differences in ECV values obtained by the different T1 estimates were small but reaching significance both in MI and remote regions (p< 0.0001). In MI, he largest difference versus MOLLI Vendor was observed for InSIL HBI(3p) with a mean difference of -4.1[-2.7;5.5] % (p< 0.001). InSiL LSQ(3p) and InSiL HBI(2p) mean differences were respectively of 1.1[1.9;0.4]% and 0.4[1.2;-0.4](p=NS). In remote regions, the largest difference of 2.1[1.7;2.5] % (p< 0.001) for InSiL HBI(2p).

Conclusion: Given the significant differences between metrics (T1 Native, T1 Gd and ECV) as a function of the post-processing algorithm used, our study pointed out the need for a vendor agnostic solution that could be systematically and retrospectively deployed for clinical trial analysis, fact checking and standardization. Not doing so, will results in maximised imprecision of markers and reduced capacity in providing fine characterization of the population and discrimination between patients and subgroups.